基于注意力和多任务的盲图像质量评价方法、装置、设备及介质
By employing an attention-based and multi-task-based blind image quality assessment method, and utilizing knowledge-guided attention modules and loss functions for supervised training, the method addresses the problem of low accuracy in existing blind image quality assessments, achieving higher assessment accuracy and generalization ability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHENZHEN UNIV
- Filing Date
- 2024-02-02
- Publication Date
- 2026-07-17
AI Technical Summary
Existing blind image quality assessment methods have low accuracy, struggle to effectively correlate different tasks, and fail to learn enough features to perform accurate blind image quality assessment.
A blind image quality assessment method based on attention and multi-task is adopted. By acquiring an image quality assessment dataset and dividing it into training and testing datasets, the attention module is guided by knowledge to learn distortion knowledge. Combined with a backbone network, an information fusion module, and an image quality score acquisition module, supervised training is performed using a first preset loss function and a second preset loss function to improve the generalization ability of the model.
It improves the accuracy of blind image quality assessment by learning sufficient features through the auxiliary task of associating distortion type with the main task of predicting image quality score, thereby enhancing the model's generalization ability.
Smart Images
Figure CN118014962B_ABST